5

Propagators and Green functions

5.1

The causality issue

Quantum field theory is buit over the same postulates as quantum mechanics, but with
fields and conjugate canonical momenta replacing positions and ordinary momenta. A key
ingredient, a priori absent in QM, is causality; i.e. the statement that information cannot
propagate faster than light.
A good way analyzing this issue is by considering the Green function or propagator G(~x, ~x 0 ; t).
This is defined as the probability amplitude fot a particle that is initially localized at point
~x to be at point ~x0 after a time t has elapsed
G(~x, ~x0 ; t) = h~x0 |eiHt |~xi.

(199)

H is assumed to be time-independent (this is almost invariably the case in field theory) and,
in fact we shall assume it is the free hamiltonian. This is also the relevant case for field theory
since perturbation theory is the analytical tool that is commonly used. Between interactions
particles are assumed to propagate freely.
It is a well known reault in QM that
G0 (~x, ~x0 ; t) = h~x0 |eiH0 t |~xi

(200)

3

m 2 im |~x0 ~x|2
2t
= d ph~x |e
|~
pih~
p|~xi =
(201)
e
2it
From this expression it is quite apparent that the propagator is non-zero bewtween any two
Z

P
i 2m
t

space-time points regardless whether one is in the causal cone of the other or not. In fact,
nothing special happens for points on the boundary of the light-cone itself.
Moving to a relativistic theory, but still sticking to the one-particle, or first quantized, interpretation, we would change (201) to
G0 (~x, ~x0 ; t) =

d3 ph~x0 |ei

p
~2 +m2 t

|~
pih~
p|~xi.

(202)

A direct calculation shows that

G0 (~x, ~x0 ; t) em

|~
x0 ~
x|2 t2

(203)

This expresion, although relativistic does not really solve the causality issue present in (201);
the propagation amplitude is small but non zero for |~x 0 ~x|2  t2 (well outside the light

cone).

36

Quantum field theory solves this problems by introducing antiparticles. We shall see that the
culprit are the negative energy modes and that causality-violating propagation of negative
energy modes is turned into causality compliant propagation thanks to antiparticles. When
we ask whether an observation made at space-time point x can influence another at y, with
(x y)2 0, we will find that the amplitudes due to particle and antiparticle propagation

exactly cancel.

To fulfill Greens equation (see below) the factor (x 0 y 0 ) is included in quantum mechanics
in the propagator.

5.2

Propagators in field theory

We shall present the discussion in terms of a scalar field and in the next section we shall
discuss the modifications for fermions and vector fields. Let us introduce the following twopoint function in field theory
D(x, y) = h0|(x)(y)|0i.

(204)

Using elementary QM techniques, this function can be written as

h0|(~x, 0)eiP

0 (x0 y 0 )

(~y , 0)|0i,

(205)

where we have defined the vacuum energy to be zero; i.e. we work with a normal ordered
hamiltonian H = P 0 . Taking into account that we go from QM to field theory by the

replacement ~x (~x), the analogy with (201) is manifest.

A simple calculation shows that D(x, y) is a c-number and, in fact, a function of x y.

h0|(x)(y)|0i =

d3 p
(2)3 2Ep

d3 k
eipx+iky h0|ap ak |0i
(2)3 2Ek

d3 p
eip(xy) .
(2)3 2Ep

(206)

(207)

Take for instance a space-like separation, x 0 = y 0 , ~r = ~x ~y , then

Z

d3 p
ei~p~r emr .
(2)3 2Ep

Again, the propagation seems to be non causal.

37

(208)

Taking into account that (x) is real field, it is actually an hermitian operation acting in the
Hilbert space of the system. Thus (x) is an observable. Now we can pose the causality issue
in very clear terms. If
[(x), (y)] = 0

(209)

a measurement made at x cannot affect another made at y. Actually we know that this is is
always the case if x0 = y 0 , as this is actually the content of one of the canonical commutation
relations. Physically it means that we can set our initial data for a constant time slice. Let
us now compute the above commutator for unequal times
[(x), (y)] =
=

d3 p
(2)3 2Ep

d3 k
[ap eipx + ap eipx , ak eikx + ak eikx ]
(2)3 2Ek

d3 p
(eip(xy) eip(yx) ) = D(x y) D(y x).
(2)3 2Ep

(210)
(211)

We see that indeed if x0 = y 0 , and changing p~ ~

p in the second integral, the commutator

is zero, as expected.

It is more interesting to note that D(x y) is a Lorentz invariant. This is not totally obvious,
but it is so by noticing that
Z

d3 p
=
(2)3 2Ep

d4 p
(p2 m2 )(p0 )
(2)3

(212)

which is manifestly invariant under proper gauge tranformations (those connected to the
identity), and that eipx is by itself invariant. Now, if (x y) 2 < 0 (y is outside the lightcone

of x, or viceversa) we can find a Lorentz transformation such that

x y (x y).

(213)

This is obviously impossible if x y is time-like, since it would reverse the time ordering of
physical events. Therefore we conclude that

if (x y)2 < 0.

[(x), (y)] = 0

(214)

Causality in the sense above indicated is now manifest. Notice that the cancellation takes
place between positive and negative energy solutions. Thus antiparticles are basic to restore
causality.
Since [(x), (y)] is a c-number, we can write
[(x), (y)] = h0|[(x), (y)]|0i.
38

(215)

Let us assume that x0 > y 0 ,

0

= (x y )

= (x0 y 0 )

(x y )h0|[(x), (y)]|0i = (x y )

d3 p
(eip(xy) eip(xy) )
(2)3 2Ep
d3 p
(2)3
d3 p
(2)3

1 ip(xy)
1
e
|p0 =Ep +
eip(xy) |p0 =Ep
2Ep
2Ep
Z

1
dp0
eip(xy) ,
2
2i p m2

where C is a circuit containing the two poles at p 0 = Ep closed through the lowest half of
the complex plane (p0 = i) since x0 y 0 > 0. The result is thus
0

DR (x y) (x y )h0|[(x), (y)]|0i =

i
d4
eip(xy) .
4
2
(2) p m2

DR (x y) = (x0 y 0 )(D(x y) D(y x)).

(216)
(217)

This is called the retarded propagator. Another way of enforcing the integration circuit is
by adding a small and negative imaginary part, declaring that the poles lie at p 0 = Ep i

and integrating p0 over the real line. Likewise we can consider an advanced propagator

with poles at p0 = Ep + i which is non-zero for y 0 > x0 , but is of no particular relevance

to us. To be precise, it is usually defined as

(y 0 x0 )h0|{(x), (y)}|0i.

(218)

We can check that the retarded propagator verifies the following differential equation
(2 + m2 )DR (x y) = i (4) (x y).

(219)

This is proposed as an exercise.

In fact all propagators are related to solutions of this partial diferential equations (Greens
y) a generic
equation) and they are therefore called Greens functions. Denoting by D(x
propagator (retarded, Feynman...)

y) = i 4 (x y),
(2 + m2 )D(x

(220)

and thus all of them in momentum space obey

(p2 + m2 )D(p)
= i,
and
y) =
D(x

i
d4 p
eip(xy)
4
2
(2) p m2
39

(221)

(222)

The difference between the different Green functions (D R (x y), DR (x y), ...) lies in the
prescription to handle the poles appearing in the p 0 integral. The retarded prescription, as

we have seen, includes both of them in the contour.

Feyman prescription amounts to considering
p2

i
.
m2 + i

(223)

Thus, poles are in p0 = Ep i and p0 = Ep + i. Then

DF =

i
d4 p
eip(xy) .
4
2
(2) p m2 + i

(224)

Now, if x0 > y 0 we have to close the circuit in the lower half-plane, and we pick only the
contribution from the pole at p0 = +Ep , that is D(x y) . The opposite happens if y 0 > x0 ;

in this case the non-zero contribution comes from the negative energy solution. In short
DF (x y) = (x0 y 0 )D(x y) + (y 0 x0 )D(y x).

(225)

h0|T (x)(y)|0i.

(226)

This is denoted as

The different prescriptions for the poles are summarized in the following figure

Advanced
Retarded
Feynman

The so-called Dyson prescription consists in taking

p2

i
m2 i

with  positive.

40

(227)

5.3

The fermion and gauge-boson propagators

The contribution from particles and antiparticles it is best discussed in a case where the
distinction matters. For this reason, and also in order to discuss the peculiarities of spin 1/2
fields.
The basic definitions are the same. Greens functions are solutions of the partial differential equation
y) = i (4) (x y).
(i 6 x m)S(x

(228)

S+ (x y) = h0|(x)(y)|0i,

(229)

S (y x) = h0|(y)(x)|0i,

(230)

SF (x y) = (x0 y 0 )h0|(x)(y)|0i
(y 0 x0 )h0|(y)(x)|0i,

(231)

We define

and

where the minus sign appears on account of the anticommuting character of the variables.
We can easily solve Greens equation in momentum space, the result is

What this actually means is

i
.
6k m

(232)

6k + m
i
=i 2
.
6k m
k m2

(233)

Feynman propagator is obtained by considering

i

k2

6k + m
.
m2 + i

(234)

An alternative way of deriving the previous expressions is by directly inserting the mode
expansion, for instance in

S+ (x y) = h0|(x)(y)|0i.

(235)

We have as a non-vanishing contribution

Z

X
d3 k
d3 p
eipx+iky u(p, )
u(k, 0 )h0|c(p, )c (k, 0 )|0i
(2)3 2Ep (2)3 2Ek ,0

X
d3 p
u(p, )
u(k, )eip(xy) .
(2)3 2Ep

41

(236)

(237)

Likewise
(x)|0i

S (x y) = h0|(y)
=

X
d3 p
v(p, )
v (p, )eip(xy) .
3
(2) 2Ep

(238)

The sum over spinors gives 6 k + m (6 k m) in the first (second) case.

We now see that h0|(x)(y)|0i

propagates particle (positive energy solutions) from y to x

while h0|(y)(x)|0i
propagates antiparticles from x to y. Causality, is now the statement

that

{(x), (y)}

(239)

vanishes for space-like separations. This can be easily proven by using the same methods as
for the bosonic case and it necessitates both both contributions, particles and antiparticles,
both propagating forward in time, but with reversed momentum.
The Feynman propagator contains both contributions too
SF (x y) = (x0 y 0 )S+ (x y) (y 0 x0 )S (y x),

(240)

but not simultaneosuly! If x0 > y 0 we have particle propagation, while for y 0 > x0 we have
antiparticle propagation.
We finally turn to gauge fields. We shall be brief in this case as we have already discussed
most of the subtle issues. Working in an arbitrary gauge, the corresponding Greens equation
is

1
(2g (1 ) )D (x y) = i (4) (x y),

whose solution is, in momentum space

(241)

g (1 ) k kk2
(i)
,
k 2 + i

(242)

where we have already adopted Feynman prescription.

5.4

Feynman interpretation of negative energy states and antiparticles

It should be clear by now that in spite of its brilliant success for spin 1/2 particles, Dirac theory
of holes and particles cannot be applied to bosons. Yet, bosons have their own antiparticles.
We are thus led to the following hypothesis due to Feynman:
The emission (absorption) of an antiparticle of 4-momentum p is physically equivalent to
the absorption (emission) of a particle of 4-momentum p
42